From bffcd8375ee8c0a15569b166c6ae593a3b800888 Mon Sep 17 00:00:00 2001 From: Wouter Besse Date: Tue, 16 May 2023 16:36:19 +0200 Subject: [PATCH] Added tybalt info to readme --- README.md | 26 +++++++++++++++++++++++++- media/Tybalt.svg | 4 ++++ 2 files changed, 29 insertions(+), 1 deletion(-) create mode 100644 media/Tybalt.svg diff --git a/README.md b/README.md index ad3eaf9..065afdb 100644 --- a/README.md +++ b/README.md @@ -45,7 +45,31 @@ example usage of CLI train.py: ### Tybalt WaveNet VAE -T.B.D. +A alteration on the [Tybalt VAE model](https://github.com/greenelab/tybalt) by [Way et al.](https://www.biorxiv.org/content/10.1101/174474v2). +I gave it one extra linear layer to help reducing the data to a smaller latent space. + +![Tybalt VAE Diagram](https://github.com/WouterBesse/Sonified-Latent-Data/blob/main/media/Tybalt.svg?raw=true) + +#### Training and model + +My model is downloadable from 'n.b.t.', it's trained on the TCGA dataset. You can train your own model using `train.py` from the WaveNetVAE folder or by using the `WaveVaePlayground.ipynb` jupyter notebook. +The acquisition and preprocessing scripts are available in the original [Tybalt GitHub](https://github.com/greenelab/tybalt). + +example usage of CLI train.py: + +`python3 train.py -tp "./traindatasetfolder/" -vp "./validationdatasetfolder/" -ep 100` +| **Short Flag** | **Long Flag** | **Description** | +|----------------|---------------------|---------------------------------------------------------------------------------------| +| `-dp` | `--data_path` | Path of folder where training data is stored, validation split is made automatically. | +| `-ep` | `--epochs` | Amount of epochs to train | +| `-ex` | `--export_path` | Path of folder to export model files to | +| `-bs` | `--batch_size` | Batch size | +| `-lr` | `--learning_rate` | Learning rate, I recommend 0.00001 | +| `-kla` | `--kl_anneal` | How much the kl rate multiplier is increased after every log step | +| `-mkl` | `--max_kl` | What the maximum kl rate multiplier will be | +| `-lpe` | `--logs_per_epoch` | How often a tensorboard log is stored per epoch | +| `-d` | `--device` | What device to train on, e.g. `cuda:0`, `cpu` | +| `-mf` | `--max_files` | The maximum amount of files to use in the train dataset | ### Mocap WaveNet VAE diff --git a/media/Tybalt.svg b/media/Tybalt.svg new file mode 100644 index 0000000..3c91f4b --- /dev/null +++ b/media/Tybalt.svg @@ -0,0 +1,4 @@ + + + +
Decoder
Decoder
Encoder
Encoder
Latent space
Latent space
Linear Block
Linear...
Linear Block (1.000)
Linear Block (1.000)
Input (5.000)
Input (5.000)
Leaky ReLU
Leaky ReLU
Batch Normalisation
Batch Normalis...
Linear
Linear
Lin. (32)
Lin. (32)
µ (32)
µ (32)
σ (32)
σ (32)
Z (32)
Z (32)
Linear (1.000)
Linear (1.000)
Linear (64)
Linear (64)
Output (5.000)
Output (5.000)
Sigmoid
Sigmoid
Sigmoid
Sigmoid
Lin. (32)
Lin. (32)Text is not SVG - cannot display \ No newline at end of file